Best of 2024 in Agents (from #1 on SWE-Bench Full, Prof. Graham Neubig of OpenHands/AllHands)

[@LatentSpaceTV]

links to slides and more context x.com/latentspacepod/status/1871998012467380698

[@RafiDude]

Please pin this comment.

[@Lucien-lu1vw]

Amazing demo with the pie charts PR. I played a lot with agents and came to the same conclusions and design choices as Graham Neubig.

[@TheNitroPython]

These talks are extremely valuable. I would love to be in person.

[@NaveenReddy-p5j]

Graham Neubig nailed it! Agent tech’s evolving fast, with AllHands on top. 2025 looks exciting for agents!

[@AlexJohnson-g4n]

AllHands leading SWE-Bench Full is impressive! Graham Neubig’s insights were top-notch. 2025 does seem like the year for major agent tech breakthroughs.

[@neuronwave]

Interesting to watch this at Christmas and reflect on how much higher o1-pro (49) and o3 (71.7) are than the performance of models on swe-bench example. Clearly highlights a challenge in 2025 (and beyond) to build things that won't be washed away in the tsunami of new models doing all the agent workbench internals. Especially relevant as the cost per hour of software dev is so high that o1-pro (and probably o3), while expensive, are much cheaper than human coder.

[@MatthewSanders-l7k]

Graham Neubig's keynote on LLM agents is a game-changer! AllHands leading SWE-Bench Full is impressive. Exciting future for agent tech!

[@WinonaNagy]

Neubig’s insights are a gem. With AI agents advancing rapidly, are we ready for a paradigm shift in our digital interactions?

[@AlternativeTakes]

can we stop ai generated answers

[@Aedonius]

Cursor IDE New YOLO MODE is insane.